Hubert-uk Open-source Model - Freely Support Ukrainian Speech-to-Text Tasks

Hubert Uk

Developed by Yehor

A Ukrainian automatic speech recognition model trained on the mHuBERT-147 base model, supporting Ukrainian speech-to-text tasks.

Speech Recognition

Safetensors

Other#Ukrainian speech recognition #Low character error rate #Efficient inference

Downloads 31

Release Time : 8/16/2024

Model Overview

This model is an automatic speech recognition (ASR) model optimized for Ukrainian, based on the mHuBERT-147 architecture and fine-tuned on Ukrainian speech datasets.

Model Features

Ukrainian Optimization

An automatic speech recognition model specifically optimized for the phonetic characteristics of Ukrainian.

Efficient Inference

Real-time factor (RTF) of 0.0026, ensuring high processing efficiency.

Multi-Metric Evaluation

Provides various evaluation metrics such as WER (37.07%) and CER (6.87%).

Model Capabilities

Ukrainian speech recognition

Speech-to-text

Use Cases

Speech Transcription

Ukrainian Speech Transcription

Convert Ukrainian speech content into text.

Word error rate 37.07%, character error rate 6.87%

Voice Assistants

Ukrainian Voice Command Recognition

Used for command recognition in Ukrainian voice assistant systems.

🚀 HuBERT for Ukrainian

A HuBERT model tailored for the Ukrainian language, enabling high - quality automatic speech recognition.

🚀 Quick Start

This section provides a quick start guide for getting the model up and running.

✨ Features

Language Support: Specifically designed for the Ukrainian language.
Community Resources: Rich community support through Discord and Telegram groups.
Evaluation Metrics: Comprehensive evaluation metrics are provided for model performance assessment.

📦 Installation

To install the necessary dependencies, follow these steps:

uv venv --python 3.12

source .venv/bin/activate

uv pip install -r requirements.txt
uv pip install -r requirements-dev.txt

📚 Documentation

Community

Discord: Join the Discord community
Speech Recognition: Telegram group for Speech Recognition
Speech Synthesis: Telegram group for Speech Synthesis

Check out other Ukrainian models: Ukrainian Speech Recognition Models

Evaluation Results

Metrics (float16) using the evaluate library:

WER: 0.3707 metric, 37.07%
CER: 0.0687 metric, 6.87%
Accuracy on words: 62.93%
Accuracy on chars: 93.13%
Inference time: 43.0227 seconds
Audio duration: 16665.5212 seconds
RTF: 0.0026

Model Information

Property	Details
Base Model	utter-project/mHuBERT-147
License	cc-by-nc-sa-4.0
Language	uk
Tags	automatic-speech-recognition
Datasets	espnet/yodas2
Metrics	wer

Model Index

Name: w2v-bert-2.0-uk-v2.1
- Results:
  - Task:
    - Name: Automatic Speech Recognition
    - Type: automatic-speech-recognition
  - Dataset:
    - Name: common_voice_10_0
    - Type: common_voice_10_0
    - Config: uk
    - Split: test
    - Args: uk
  - Metrics:
    - Name: WER
    - Type: wer
    - Value: 37.07
    - Name: CER
    - Type: cer
    - Value: 6.87

📄 License

This project is licensed under the cc-by-nc-sa-4.0 license.

📚 Cite this work

If you use this work in your research, please cite it as follows:

@misc {smoliakov_2025,
	author       = { {Smoliakov} },
	title        = { hubert-uk (Revision 4aae976) },
	year         = 2025,
	url          = { https://huggingface.co/Yehor/hubert-uk },
	doi          = { 10.57967/hf/4557 },
	publisher    = { Hugging Face }
}

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご